Iris Train S Usr Rewrite: 100K LLM Instruction Rewrites for TOTE-IR Training

100,000 rows of rewritten user instructions for training or evaluating instruction-following language models. The data was created by author ggwplarin and involves LLM rewrites of the user/context section, with system prompts intentionally omitted. This dataset was last updated on May 23, 2026.

Use Cases

• Training instruction-following models based on rewritten user prompts.
• Analyzing the impact of user instruction rewrites on target intermediate representation generation.
• Benchmarking LLM rewriting models using the provided metadata.
• Studying source lineage and data transformation patterns for synthetic training data creation.

Strengths

• Contains 100,000 rows of rewritten user instructions, providing a substantial corpus.
• Includes source lineage fields and rewrite model metadata, which may aid in reproducibility and analysis.
• Focuses specifically on user/context rewrites with system prompts omitted, creating a targeted training resource.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is the only known scale metric; file formats and exact column definitions are unknown.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

LLM rewriting of user/context sections from a source accepted parquet file; inventory, bindings, and target IR sourced from the original.

Freshness

Last updated 2026-05-23 09:48:24; freshness should be verified.